Neuromodulated Dopamine Plastic Networks for Heterogeneous Transfer Learning with Hebbian Principle

The plastic modifications in synaptic connectivity is primarily from changes triggered by neuromodulated dopamine signals. These activities are controlled neuromodulation, which itself under the control of brain. subjective brain’s self-modifying abilities play an essential role learning and adaptation. artificial neural networks with plasticity used to implement transfer image classification domain. In particular, this has application detection, segmentation, parameters significant results. This paper proposes a novel approach enhance accuracy heterogeneous source target, using neuromodulation Hebbian principle, called NDHTL (Neuromodulated Dopamine Transfer Learning). Neuromodulation offers powerful new technique applications training implementing asymmetric backpropagation principles motivated CNNs (Convolutional networks). Biologically concomitant learning, where connected brain cells activate positively, enhances connection strength between network neurons. Using algorithm, percentage change neurons CNN layer directly managed signal’s value. discriminative nature fits well technique. learned model’s weights must adapt unseen target datasets least cost effort learning. distinctive such as for gradient update approach. emphasizes algorithmic signals classify images source-target datasets. standard symmetric framework. Experimental results CIFAR-10 CIFAR-100 show that proposed algorithm can efficiency compared existing methods.